Topic: Proof of Useful Work? (Read 584 times)

Proof of work problems need to satisfy several properties. They need to be quick to verify, hard to solve, have a fine-tunable difficulty level. They also need to be progress free (a miner that started 5 minutes ago but who did not achieve a block will have no advantage over someone who started 1 minute ago). The solution to the proof of work problem also needs to be tied to the blockchain to prevent reuse of solutions and precomputing solutions. Finally, it should not be possible for one to construct a secret algorithm for solving the proof of work problem that is significantly better than the standard algorithm (ahem, asicboost and possibly approximate mining; fortunately these shortcuts were not bad enough to ruin Bitcoin). These conditions that proof of work problems must satisfy make it much harder to make a more useful proof of work problem. And to make things a bit worse, the proof of work must be useful for the public, but not for the miners themselves for reasons that people have already laid out in this thread.

With this in mind, it is probably best to think about what sort of energy efficient computing technology we will need to use in the far future but is hard to develop now because a lack of a market and then design a mining algorithm to incentivize the development of that energy efficient future computing technology. Please read the recent paper by Ralph Merkle on energy efficient mechanical computation to get an idea of what kind of energy efficient computing technology we should try to incentivize.

This, people see PoW as flawed because it "waste" computational power to solve the SHA256 puzzle but they miss the point. This energy is not wasted it is used to secure the network and makes it easily auditable. Eg: you know roughly how many watts of energy you need for a round of SHA256 which can be easily extrapolated to the network. This gives an ability to easily audit and verify nothing was tempered with.

There are something similar to what you described which is boinc distributed computer. You use your GPU for computation and get rewarded in "points" replace those points by token and you have something similar. However right now those point are worthless because they are just a way of ranking and you have no way to easily audit (by non mining node).
* How do you verify the work was done?
* In boinc it is centralized which means effectively you could print money

The main problem with this idea I believe is that SOMEONE must decide what problems to give to 'the work' being done. Bitcoin PoW works because it is just general work being done, and it is useful, not wasted, because it secures the blockchain, which is obviously of prime importance.

If you want to have a system in which the work gets applied to some outside series of computational problems then someone has to be inputting these problems into the system. I mean I guess you could have the users vote on the upcoming problems to be worked on or something, but the whole design seems rather flimsy and like it might work better for a very centralized network but not for a global open decentralized system like Bitcoin.

Then again, no cryptos are actually trying to compete with Bitcoin so the crypto ecosystem is, other than Bitcoin, made up of fairly centralized networks. But still, I think this idea of PoUW is flimsy even for altcoins for the most part. Maybe for an XRP-like network, which it is super centralized and a single entity is the central authority and acts as gatekeeper for all nodes, then it could work because that single entity would also be deciding on the work to be done so all nodes would have to comply due to the focus on centralization. But that sort of system is already anethema to the entire idea of crypto so it seems unlikely to succeed in the market just trying to ride off this idea of 'useful' work.

Not entirely. ASICs made mining more efficient, but it’s the honest entities/good actors that take the costs to spend energy to find the appropriate Proof of Work that prevent 50% attacks. Mining can be done without ASICs.

Yes, it is somewhat correct to say that. The ASIC manufacturers have a distinct financial incentive for the miners to not harm bitcoin.

If the same test set is continuously used, the resulting model will simply overfit the test data (it would obviously need to be public), and would likely not perform well against data "in the wild" -- it would perform poorly in production.

Thank you.
Good one to know. So it is correct to say that the manufacturers of ASICs are the people indirectly preventing 51% attack.

I thought about a system based on solving neural network i think all these sort of solution will have the same kind of issue.

With a neural network you could have this principale of pow with a learning set and shorter testing set the goal of solving the problem would be finding a network configuration that have lower error ratio than the previous on the testing set. Then the one who found the solution gain the reward. You could probably even make in sort that an identifer associated with a public key is "laced" into the computation in sort that you cant "steal" the work.

Seem the algorithm in the paper would use zero knowledge proof for this.

But the first problem is there will not neccesarily be a match between the demand for a particular problem solving and the amount of computation required to match a constant block time with a given amount of computational power.

In that case you have a reward for solving a particular problem and thats the economy of it.

If it takes two month to find the useful solution, there will not be any reward emitted for 2 month. But in this context of problem solving its fine, the economic value of the network is based on the amount of problem it can solve, so if there is a reward if 1000$ to solve this particular problem, and it take 2 month with the available computing power, then after the two month the network has increased its value of 1000$.

But its a completely different economy than emitting a currency in a p2p network. It an economy of decentralised cloud computing.

Beside this the only kind of work that can really be fit optimally for the purpose of emitting a coin is something approaching a random computation because it can evenly dustribute reward among all participant based on pure computing power.

It seems the paper address the issue of bounded time computation but to me it seems its more in that the ratio of verifier vs solver is always going to be low, more than to garantee a maximum time boundary for solving the problem ( doesnt seem very clear to me ).

But in the economy of cloud computing for solving problem its not really the same constraint.

To me it seem a bit pointless to try to mix the two because its very unlikely you will always have a perfect match between the demand for computation in problem solving and the problematic for a distributed cash system.

Energy expanditure is universal enough to be a good economic marker, computational problem solving much less.

You are correct in your understanding.

There are methods of mining, other than PoW. One of which is Proof of Stake (PoS). From a security standpoint, it is my belief that PoW is more secure because PoW has a more extensive incentive structure. For example, PoS requires the "miners" to hold a certain amount of the coin they are mining, and the frequency any particular miner will produce a block is proportional to the amount of coin the miner is holding. It is trivial for someone to acquire an arbitrary amount of a PoS coin on the open market. However, with bitcoin's PoW, someone who wants to mine bitcoin will need to purchase an ASIC from a mining manufacturer (or from someone who currently owns an ASIC on the secondary market, with the owner ultimately purchased the ASIC from the manufacturer). The mining manufactures have large upfront R&D costs associated with manufacturing ASICs. So the mining manufacturers have an incentive to not sell large quantities of miners to a single entity who could potentially do harm to the bitcoin network, or else they are risking their R&D investments.

Thank you for response. So it means that the energy expended is not used to add transaction to the next block but to secure the transaction. So, all the miners commit energy to the system to secure it. If the system fails their cost of energy will waste.

Work performed via bitcoin's PoW does not accomplish the task of inserting a transaction into a block.

The work performed via bitcoin's PoW guarantees the order of blocks. If someone wanted to create a block that competes with the most recent block, assuming they have the necessary equipment, they would have to "pay" in the form of the excess energy expanded required to create a competing block. So the PoW essentially guarantees the security of a bitcoin transaction.

It’s going to work if the “work” is also used as a Sybil Attack prevention mechanism for the network. Because what would be the actual point of POW for the network? I have seen posts about Primecoin as “useful POW” when I was doing research on different mining algorithms.

This is a topic that comes up again and again over the years: algorithms that 'mine' and at the same time find large prime numbers, hardware made to heat your home while mining; all sorts of ideas.
In the end, you can attach a monetary value to everything, so what you're doing with 'useful work' ideas is just increasing the miner's efficiency.

For example, if you can extract $100 of heat and $100 of BTC opposed to just the BTC reward, you've essentially just doubled the device's efficiency. Same goes for finding a $100 worth large prime and $100 worth of BTC. Sounds great. However, this is not where the story ends. The network will have one (or many - others will start doing this same thing) way more efficient miner, which will cause shorter block times and a difficulty adjustment proportional to the efficiency increase.

In the long run, the primes or heat costs will get cheaper (for example) and together with the diff adjustment, in theory you will end up with $100 worth of revenue again, just like it started before all those efficiency optimizations.

I have feeling that the people that wrote this paper are hiding some loopholes under technicality. It is an excellent idea but the road map is the same thing as the PoW. Where it was suppose to find another route it did not do it clearly.
I expected to see a clear computing steps which will convert the inputs to useful outputs by maintaining hardness and usefulness.
I saw in the document where they said.
Does the above correctly describe bitcoin PoW?

It is not possible that an oracle could be truly random and decentralized at the same time.

And the concerns raised in the comments seem to be right. If solving mathematical problems is profitable here, is PoUW not merged mining to some extent? Allowing consequently "zero cost attacks", or something very similar to zero cost attacks?

The Bitcoin protocol is not very well suited to the needs of merged mining. Much more than just the algo would have to be changed, and even then the current level of security is unlikely to be maintained.

Merged mining (the real one at least) can be and is a great solution, but for other coins with a different purpose than Bitcoin.

Any kind of "work" is going to have value.

With bitcoin's current incentive structure, a miner will perform PoW "for" bitcoin users, who in turn "pay" the miner in the form of the block subsidy. Miners will also confirm transactions for a specific subset of bitcoin users, who will also pay the miners in the form of transaction fees. ASICs used to mine bitcoin cannot be used to perform any other valuable work that comes close to the value of performing PoW "for" bitcoin users (performing PoW with the hopes of finding a block), so if the miners as a whole were to do something detrimental such as routinely accept double-spend transactions in a block race, or perform a 51% attack against bitcoin, the value of the ASIC mining equipment will decline to nearly zero.

On the other hand, if a bitcoin miner were performing some kind of PoW to try to find a bitcoin block (for the purpose of receiving the block reward -- that is the block subsidy plus tx fees), in addition to performing additional "useful" work that is valuable to someone else, the miners will not have the same incentives to not attack bitcoin. If the value of the other work is far in excess of the value of the EV of the block reward, a miner may seek short-term profit in attacking bitcoin, or they may not even bother with mining bitcoin. If the miners were to attack bitcoin, the value of their equipment would not drop to near zero because revenue could still be obtained via performing the other "useful" work.

A public good is typically paid for by the government. If a government cannot obtain a public good via its own resources (for example, a country that does not have the manufacturing capacity to produce tanks), will seek to exchange its abundant resources (which may include money) for resources that will produce said public good. Further, if a government cannot provide a public good period, its wealthy citizens will seek to use their own resources to obtain said public good amongst themselves (for example, a neighborhood building a security wall and hiring private security when the local government cannot handle crime). So there will always be someone willing to pay for a public good, and if you can provide a public good, you should expect to receive compensation for providing this service.

No mechanism needs to be changed from any crypto. The point of this thread is to examine if Proof of Useful work can actually work in practise.

“Should be seen as useful”? For what, and whose benefit? If people cannot, or do not want to understand how the protocol works, then the ignorance is on them. POW doesn’t need to change just “to be seen as useful”.

Yes, I did, even the paper. I am sorry if I didn’t write clearly enough- I think that ir.hn described well what I wanted to say - that work should be seen as a useful, if there is so much waste of energy just to calculate hash, many people feel uncomfortable with that. Especially that from time to time we see comparisons in newspapers “bitcoin is using as much energy as Paraguay” or so. And then argument that it is cost of keeping the system alive is not enough.

Just a small thought. What if we only change the task, from current one (non productive) to something productive?
I would think about something like SETI or (of which I am member) Biggest prime search - mersenne.org
Computer may receive some tasks to compute and I believe it is still possible to estimate their difficulty and expected time to process - so rule of 10 minutes may be kept.

It sounds like another idea from the Space Ghost MEME who just discovered Bitcoin, and he’s here to “fix it”.

The truth is POW makes the incentive structure possible, which is the structure that keeps everything in the network to stick together. It solves the Byzantine General’s Problem, it’s protecting it from being Sybillized and it’s responsible for the issuance of the currency. Bitcoin is a closed working loop because of POW.

POW as implemented in Bitcoin is useful.